1. Verification and Performance Estimation Environment for 3D Graphics Geometry Acceleration System 995022 Young-Su Kwon

2. Presentation Plan 1st presentation Motivation of 3D graphics system design(FGA system) Why cosimulation environment is needed? The various methods of HW/SW systems and our method 2nd presentation Previous works for Performance estimation Problems of previous works ( Simulation performance, The efficiency of abstraction level ) The appropriate method for 3D graphics system design 3rd presentation FGA system cosimulation environment 4th presentation Previous methods for design exploration FGA system exploration ( library development and performance improvement) 5th presentation System performance improvement according to system design exploration 1st presentation Motivation of 3D graphics system design(FGA system) Why cosimulation environment is needed? Cosimulation examples and our environment

3. Motivation o The increased importance of “Geometry stage” in 3D graphics used in games, CAD, and WWW browsers run on PCs. 12.1%81.2%6.7% 13.4%82.0%4.6% OtherGeometry stage Rendering stage Time distribution of 3D graphics application on Pentium II 250MHz and Voodoo2 gears reflect

4. Geometry Stage Acceleration o Approaches for geometry acceleration u 3D acceleration features on host CPU u Independent acceleration system(board) o Additional features on host CPU u AMD 3D Now! u Pentium III ( Streaming SIMD insrucitons) u SIMD FPU instructions o Geometry acceleration system u FGA system u Fujitsu & Sony? ( PlayStation )

5. FGA(FLOVA Geometry Accelerator) Graphics System o FGA graphics system u Geometry accelerated 3D graphics system u Host + FGA + Rendering Engine Host (Pentium) running FGA-GL Voodoo2 FLOVA ROM RAM 4-way VLIW Processor with Floating-point units Newly on PC add-on slot for geometry processing FGA PCI

6. Co-simulation of FGA and FGA-GL o No FLOVA chip(No FGA, i.e, No HW) o Pre-design(functional verification) of FGA-GL(SW part) and FGA(HW part) before chip fabrication(Development time requirement) o Earlier performance estimation and FGA-GL optimization to satisfy performance requirement (3.0-4.0 performance improvement for 3D benchmarks without FGA )

7. Microprocessor HW/SW Cosimulation o Cosimulation examples o Microprocessor systems for specific application u Some portion of process is executed in HW. Application code I/O drivers Interface & HW o Design of ASIP u Specific application u HW/SW partitioning is important. Application code SW HW Controller Datapath SW HW o Design of Co-processor Application code Controller Datapath SW HW Microprocessor

8. FGA Cosimulation Environment Rendering (Vertex rendering to frame buffer) Geometry Processing (Vertex coordinates and color calculation) Context Setting HSM (Hardware Simulation Model) Virtual Memory ISS (VLIW processor model) API Voodoo2 SSM (Software Simulation Model)

9. Summary o Motivation of 3D graphics geometry acceleration system development o FGA system configuration o The need of cosimulation environment o Cosimulation examples and FGA system cosimulation environment

10. Performance Estimation Architecture selection (Definition of API interface) Classification of timing-cosuming part Time measurement of each part Real application program running SSM core geometry part context setting part rendering part Transaction HSM kernel running part

11. App. Area System Operation glBegin() glEnd() VB(i) filling glBegin() glEnd() Vb(i+1) filling App. Area FGA comm. Transformed VB(i-1) & context VB(i) Transformed VB(i) & context VB(i+1) Geometry processing on VB(i) FGA comm. Geometry processing on VB(i-1) FGA comm. Geometry processing context i context i+1 RE control context i context i+1 context i context i-1

12. Pitfall u glBegin()/glEnd() 사이에서 Vertex filling 을 하 던 중 FGA 용 VB size 를 초과했을 때  New glBegin/End 로 assume. App. Area glBegin() VB(i) filling glEnd() Vb(i+1) filling FGA comm. VB(i) Geometry processing on VB(i) FGA comm. Geometry processing on VB(i-1) context i RE control context i